Kevin Zakka's banner

Kevin Zakka

@kevin_zakka • 19,820 subscribers

phding @Berkeley_AI

Shorts

Booster recovery controller from last night. Sim design, training and deployment on hardware took < 1 day. With Qiayuan Liao

Booster recovery controller from last night. Sim design, training and deployment on hardware took < 1 day. With Qiayuan Liao

104,613 görüntüleme

Proprioception only

Proprioception only

45,574 görüntüleme

Ported over the getup tasks from playground to mjlab for my thesis defense. Will make them available in a separate repo so others can repro and build on top in the coming weeks.

Ported over the getup tasks from playground to mjlab for my thesis defense. Will make them available in a separate repo so others can repro and build on top in the coming weeks.

24,836 görüntüleme

The MuJoCo team just released a touch sensor using the new plugin API. It essentially aggregates all contact forces into an array of rectangular pixels to create an image. Here's a braille demo from the amazing Yuval Tassa:

The MuJoCo team just released a touch sensor using the new plugin API. It essentially aggregates all contact forces into an array of rectangular pixels to create an image. Here's a braille demo from the amazing Yuval Tassa:

75,919 görüntüleme

MuJoCo 3.0 is out and it comes packed with amazing new features: TPU/GPU acceleration, non-convex geometries and deformable bodies!

MuJoCo 3.0 is out and it comes packed with amazing new features: TPU/GPU acceleration, non-convex geometries and deformable bodies!

57,007 görüntüleme

Happy Monday! We just added raycast sensors to mjlab. They let a robot sample the scene with virtual rays, returning distances and surface geometry. Rays can use orthographic projection for terrain scanning or perspective projection for simulating a depth camera.

Happy Monday! We just added raycast sensors to mjlab. They let a robot sample the scene with virtual rays, returning distances and surface geometry. Rays can use orthographic projection for terrain scanning or perspective projection for simulating a depth camera.

20,105 görüntüleme

A lot of people have been asking so I made a repo to show how easy it is to setup a custom robot in mjlab. You don't need to clone and copy your robot into mjlab!

A lot of people have been asking so I made a repo to show how easy it is to setup a custom robot in mjlab. You don't need to clone and copy your robot into mjlab!

16,511 görüntüleme

Demo starting in 10 minutes, come witness the magic of open-source sim2real!

Demo starting in 10 minutes, come witness the magic of open-source sim2real!

20,992 görüntüleme

MuJoCo 3.1.6 has just been released! MJX now comes with out-of-the-box support for height fields. Check out this joystick policy that Baruch Tabanpour trained (~2M steps per second on an A100). Colab to reproduce:

MuJoCo 3.1.6 has just been released! MJX now comes with out-of-the-box support for height fields. Check out this joystick policy that Baruch Tabanpour trained (~2M steps per second on an A100). Colab to reproduce:

30,957 görüntüleme

Whole body retargeting is dead simple with 𝗺𝗶𝗻𝗸 and runs much faster than real time. Here is an example of an AMASS sequence retargeted to Unitree's H1 humanoid. Kudos to Arthur Allshire for helping with this 🙂

Whole body retargeting is dead simple with 𝗺𝗶𝗻𝗸 and runs much faster than real time. Here is an example of an AMASS sequence retargeted to Unitree's H1 humanoid. Kudos to Arthur Allshire for helping with this 🙂

22,503 görüntüleme

Just released a MuJoCo implementation of end-effector control with differential inverse kinematics. It's blazing fast, super readable, has 0 external dependencies and best of all, it's just a few lines of code:

Just released a MuJoCo implementation of end-effector control with differential inverse kinematics. It's blazing fast, super readable, has 0 external dependencies and best of all, it's just a few lines of code:

26,427 görüntüleme

mink 1.0.0 is out! You can now enforce tasks as strict QP equalities. One neat application is "locking" certain DoFs. In this example, joints 1 & 2 of the Panda are locked when the cube is red and then unlocked when it turns green.

mink 1.0.0 is out! You can now enforce tasks as strict QP equalities. One neat application is "locking" certain DoFs. In this example, joints 1 & 2 of the Panda are locked when the cube is red and then unlocked when it turns green.

10,186 görüntüleme

Just added an example of how to implement a simple gaze tracking task on the Apptronik robot in mink:

Just added an example of how to implement a simple gaze tracking task on the Apptronik robot in mink:

15,707 görüntüleme

Videos

Anya Rossi

sweetdream.ai

SweetDream.ai•Sponsored•Livecam

Watch Anya Live

Anya is streaming live right now! Join her private show and enjoy exclusive content.

Exclusive private shows

1.2k viewers online

Private Show

Join now for exclusive access

Free preview available • Premium content

20 years ago

20 years ago

335,281 görüntüleme • 2 yıl önce

The ultimate test of any physics simulator is its ability to deliver real-world results. With MuJoCo Playground, we’ve combined the very best: MuJoCo’s rich and thriving ecosystem, massively parallel GPU-accelerated simulation, and real-world results across a diverse range of robot platforms: quadrupeds, humanoids, dexterous hands, and arms. Best of all? You can get started today with a single command: pip install playground

The ultimate test of any physics simulator is its ability to deliver real-world results. With MuJoCo Playground, we’ve combined the very best: MuJoCo’s rich and thriving ecosystem, massively parallel GPU-accelerated simulation, and real-world results across a diverse range of robot platforms: quadrupeds, humanoids, dexterous hands, and arms. Best of all? You can get started today with a single command: pip install playground

154,212 görüntüleme • 1 yıl önce

Introducing 𝗥𝗼𝗯𝗼𝗣𝗶𝗮𝗻𝗶𝘀𝘁 🎹🤖, a new benchmark for high-dimensional robot control! Solving it requires mastering the piano with two anthropomorphic hands. This has been one year in the making, and I couldn’t be happier to release it today! Some highlights below:

Introducing 𝗥𝗼𝗯𝗼𝗣𝗶𝗮𝗻𝗶𝘀𝘁 🎹🤖, a new benchmark for high-dimensional robot control! Solving it requires mastering the piano with two anthropomorphic hands. This has been one year in the making, and I couldn’t be happier to release it today! Some highlights below:

260,159 görüntüleme • 3 yıl önce

I'm super excited to announce mjlab today! mjlab = Isaac Lab's APIs + best-in-class MuJoCo physics + massively parallel GPU acceleration Built directly on MuJoCo Warp with the abstractions you love.

I'm super excited to announce mjlab today! mjlab = Isaac Lab's APIs + best-in-class MuJoCo physics + massively parallel GPU acceleration Built directly on MuJoCo Warp with the abstractions you love.

92,433 görüntüleme • 1 yıl önce

Booster recovery controller from last night. Sim design, training and deployment on hardware took < 1 day. With Qiayuan Liao

Booster recovery controller from last night. Sim design, training and deployment on hardware took < 1 day. With Qiayuan Liao

104,613 görüntüleme • 1 yıl önce

Coming to mjlab today! This is vanilla RL, no motion imitation/AMP. Natural gaits emerge from minimal rewards: velocity tracking, upright torso, speed-adaptive joint regularization, and contact quality (foot clearance, slip, soft landings). No reference trajectories or gait patterns. Walking, running, and arm swing emerge purely from optimizing these simple objectives. Oh and training time? Just 1 hour.

Coming to mjlab today! This is vanilla RL, no motion imitation/AMP. Natural gaits emerge from minimal rewards: velocity tracking, upright torso, speed-adaptive joint regularization, and contact quality (foot clearance, slip, soft landings). No reference trajectories or gait patterns. Walking, running, and arm swing emerge purely from optimizing these simple objectives. Oh and training time? Just 1 hour.

64,807 görüntüleme • 11 ay önce

IK alone can take you pretty far. I put together a really cute demo to demonstrate this. Watch this Panda robot CNC engrave the MuJoCo logo into a curved dome using mink.

25,546 görüntüleme • 3 ay önce

I trained a grasping policy on objects of various shapes and sizes using the new mjlab feature and this really cool pivot grasp strategy emerges for large flat objects that are larger than the gripper aperture. So cool and beautiful to see!

I trained a grasping policy on objects of various shapes and sizes using the new mjlab feature and this really cool pivot grasp strategy emerges for large flat objects that are larger than the gripper aperture. So cool and beautiful to see!

20,458 görüntüleme • 5 ay önce

Really excited to release mjviser, a web-based MuJoCo viewer, powered by Viser. It has almost all the features of the native MuJoCo viewer, but runs in your browser. Load and simulate any MuJoCo model with a single uv command 👇 uvx mjviser

Really excited to release mjviser, a web-based MuJoCo viewer, powered by Viser. It has almost all the features of the native MuJoCo viewer, but runs in your browser. Load and simulate any MuJoCo model with a single uv command 👇 uvx mjviser

22,907 görüntüleme • 6 ay önce

Of course mjlab supports the native MuJoCo viewer. Makes it a breeze to pause, slow down, inspect contacts, perturb the robot, etc. There's also a brand new pane for reward visualization :)

32,318 görüntüleme • 1 yıl önce

Meet mjlab. Powered by MuJoCo Warp. Drops Monday.

Meet mjlab. Powered by MuJoCo Warp. Drops Monday.

25,017 görüntüleme • 1 yıl önce

Just finished implementing collision avoidance in the diff IK solver as extra constraints in the QP. Really happy with the final result 😀

Just finished implementing collision avoidance in the diff IK solver as extra constraints in the QP. Really happy with the final result 😀

41,968 görüntüleme • 2 yıl önce

New in mjlab from the amazing Takahiro Miki: 8 new terrains and a viser-based terrain visualizer 😎

New in mjlab from the amazing Takahiro Miki: 8 new terrains and a viser-based terrain visualizer 😎

17,050 görüntüleme • 8 ay önce

Differential inverse kinematics controller on a Go1 quadruped regulating the pose of the root body and the position of the feet.

Differential inverse kinematics controller on a Go1 quadruped regulating the pose of the root body and the position of the feet.

38,590 görüntüleme • 2 yıl önce

We've just added the Booster Robotics T1 Humanoid to MuJoCo Playground, with code to train a joystick controller as well as a deployable ONNX file. A super fun weekend with Carlo Sferrazza and Qiayuan Liao :)

We've just added the Booster Robotics T1 Humanoid to MuJoCo Playground, with code to train a joystick controller as well as a deployable ONNX file. A super fun weekend with Carlo Sferrazza and Qiayuan Liao :)

29,927 görüntüleme • 1 yıl önce

Coming soon to mjlab and a long time in the making: RGB-D camera rendering! We can solve cube lifting with the YAM arm from 32×32 RGB frames in about <5 minutes of wall-clock time. Here's a clip showing emergent "search" behavior along with our upcoming viser visualization.

Coming soon to mjlab and a long time in the making: RGB-D camera rendering! We can solve cube lifting with the YAM arm from 32×32 RGB frames in about <5 minutes of wall-clock time. Here's a clip showing emergent "search" behavior along with our upcoming viser visualization.

18,317 görüntüleme • 10 ay önce

Ever wanted to simulate an entire house in MuJoCo or a very cluttered kitchen? Well now you can with the newly introduced sleeping islands: groups of stationary bodies that drop out of the physics pipeline until disturbed. Check out Yuval's amazing video and documentation 👇

Ever wanted to simulate an entire house in MuJoCo or a very cluttered kitchen? Well now you can with the newly introduced sleeping islands: groups of stationary bodies that drop out of the physics pipeline until disturbed. Check out Yuval's amazing video and documentation 👇

18,740 görüntüleme • 11 ay önce

Center-of-mass regulation + feet/hands on the G1 humanoid.

Center-of-mass regulation + feet/hands on the G1 humanoid.

30,070 görüntüleme • 2 yıl önce

mjlab v1.3.0 is out! It's packed with features: a revamped viewer built on mjviser, a preset-based terrain system, simplified actuator config, and new MDP primitives. Here's a Go1 quadruped trained in the stair terrain with the new terrain-aware foot height sensor.

mjlab v1.3.0 is out! It's packed with features: a revamped viewer built on mjviser, a preset-based terrain system, simplified actuator config, and new MDP primitives. Here's a Go1 quadruped trained in the stair terrain with the new terrain-aware foot height sensor.

10,580 görüntüleme • 6 ay önce

1957 robot teleop

1957 robot teleop

34,339 görüntüleme • 3 yıl önce